HAL Id: hal-01211017
https://hal.archives-ouvertes.fr/hal-01211017
Submitted on 3 Jun 2020
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Accuracy of the FeedPhone device for recording eating and rumination times in dairy cows
Remy Delagarde, Jean-Pierre Lemonnier
To cite this version:
Remy Delagarde, Jean-Pierre Lemonnier. Accuracy of the FeedPhone device for recording eating and
rumination times in dairy cows. 18. Symposium of the european grassland federation, Jun 2015,
Wageningen, Netherlands. Academic Publishers, Grassland Science in Europe, 20, 2015, Grassland
Science in Europe. �hal-01211017�
Grassland Science in Europe, Vol. 20 – Grassland and forages in high output dairy farming systems 203
Accuracy of the FeedPhone device for recording eating and rumination times in dairy cows
Delagarde R. 1 and Lemonnier J.P. 2
1 INRA, UMR1348 INRA-Agrocampus Ouest PEGASE, 35590 Saint-Gilles, France; 2 MEDRIA, CS 10031 Châteaubourg, 35538 Noyal-sur-Vilaine, France
Abstract
Several commercially available devices automatically record feeding behaviour of dairy cows on farm, but independent validation studies are often not available. The objective of this study was to determine the accuracy of the FeedPhone® device, developed in France by Medria, to record eating and rumination activities of dairy cows. The FeedPhone is based on a tri-axial accelerometer placed on a collar, data being radio-transferred and processed automatically. The main activity (eating, rumination, or rest) is recorded every 5 min. Validation was performed on 7 lactating dairy cows fed on maize silage and concentrates for a total of 89 full day records. The actual times were determined by a reference method, by recording continuously the weight of the trough and the jaw movements at the minute scale. At the day level, the mean prediction error was 11.5% for eating time and 11.1% for rumination times, with low mean and slope biases (error mainly random). Eating and rumination activities are clearly distinguishable. This precision enables the detection of between-day variations of both eating or rumination times of 20, 10, and 5%, at cow level, small-herd level (4-7 cows) and larger herd level (>20 cows), respectively. This accuracy makes the FeedPhone valuable for studying relative variations of both eating and rumination times of dairy cows fed on total mixed ration.
Keywords: cattle, behaviour, methodology, accelerometer, accuracy
Introduction
Ruminant feeding behaviour has been recorded for a long time for scientific purposes to better understand and predict animal-feed relationships. More recently, advanced technologies allow continuous recording of on-farm dairy cow feeding behaviour and oestrus activity while also detecting health or nutritional events such as acidosis periods, either at cow or herd levels. This involves, however, accurate recording of eating and/or rumination activities. Several devices have been developed recently, most of them being based on biaxial or tri-axial accelerometers (Nielsen, 2013; Umemura, 2013). The objective of this work was to determine the precision and accuracy of the FeedPhone®, a new device for automatically recording eating and rumination times and the daily pattern of these events, which can be used on-farm for dairy herd management but also in ruminant research studies.
Materials and methods
Seven mid-lactating Holstein dairy cows were fed ad libitum during several weeks on a total mixed ration
which comprised 59% maize silage, 39% concentrates and 2% minerals. After several days of adaptation,
feeding behaviour was recorded continuously during 20 days from December 2013 to January 2014 by
two independent methods: (1) the INRA reference method for measuring actual eating and rumination
times, and (2) the Feedphone device developed by Medria for measuring predicted eating and rumination
times. The INRA method consists of combining data from automatic weighing troughs and from bite
meters. The bite meter is made up of a small balloon filled with foam rubber placed on the lower jaw and
connected by a silicone tube to a pressure recorder (Baumont et al., 2004). Feeding activity is interpreted
each minute and classified as eating, rumination, or resting. The FeedPhone uses the Axel® sensor placed
on a properly fitted collar on the cow’s neck. The sensor consists of a micro-electromechanical tri-axial
accelerometer that measures and continuously analyses the changes of inclination and lateral and vertical
accelerations. A set of nine statistical data is recorded every 5 min, automatically transmitted to the Box Medria by a radio link and thereafter to the data centre by a GPRS radio or internet connection. In the data centre, the processing algorithms on servers convert the raw data into new standardised data. The algorithms determine the main activity (here also classified as rumination, eating, or rest) for every 5 min period. They also determine if the animal is standing or lying down, and distinguish between eating at the barn or at grazing.
The accuracy of the FeedPhone in estimating eating and rumination time was investigated by calculating the mean prediction error (MPE, in min or in proportion to actual mean value), and the proportional contribution of mean bias, line bias and random variation to the mean-squared prediction error (MSPE) (Bibby and Toutenburg, 1977). Accuracy was investigated per day or per hour, by summing activities recorded at 5 min periods (FeedPhone) or at every 1 min period (INRA). Short unclear behaviour periods from the INRA reference system were not considered.
Results and discussion
The validation study involved a total of 89 complete cow × day recordings, for a total of 2,212 h of validation. At the day scale, the mean prediction error was 11.5% (42 min d -1 ) for eating time and 11.1%
(59 min d -1 ) for rumination time, when the FeedPhone is compared to the INRA reference method used as the ‘gold standard’. As no ‘gold standard’ is perfect (the INRA method is also an indirect method), it may be hypothesized that true accuracy should probably be greater. For eating time, part of the error comes from an under-estimation of low actual eating times (Figure 1), leading to a small overall under- estimation of eating time (-18 min d -1 , 18% of MSPE; Table 1). For rumination time, error is mainly random (97% of MSPE) with low mean and slope biases but greater dispersion (Figure 1). At the hour scale, mean prediction error is of 39% (5.7 min h -1 ) for eating time and 35% (7.5 min h -1 ) for rumination
FeedPhone (min d
-1)
0 100 200 300 400 500 600
INRA (min d-1)
0 100 200 300 400 500 600
FeedPhone (min d
-1)
200 300 400 500 600 700 800
200 300 400 500 600 700
Eating time 800 Rumination time
Figure 1. Daily eating and rumination times estimated by the FeedPhone or by INRA reference method on 7 cows during 20 days (total of 89 recordings).
Rumination time (min h
-1)
0 10 20 30 40 50 60
Bi as (F eed Phone m inu s I NR A)
-60 -40 -20 0 20 40 60
Eating time (min h
-1)
0 10 20 30 40 50 60
-60 -40 -20 0 20 40
Eating time (min h
-1) 60 Rumination time (min h
-1)
Figure 2. Relationship, for each of the 2,212 h of validation, between the bias (FeedPhone minus INRA) in eating or rumination time and the
other feeding activity time (rumination for eating, and eating for rumination).
Grassland Science in Europe, Vol. 20 – Grassland and forages in high output dairy farming systems 205 time, with no mean or slope biases, 97% of MSPE being random (Table 1). At the hour scale, there was no significant relationship between eating time bias and rumination time, nor between rumination time bias and eating time (Figure 2). It is noticeable that when a cow is engaged in a feeding activity (>40 min h -1 ), there are only very small biases in the other feeding activity (Figure 2). This clearly shows the ability of the processing algorithms to classify feeding activities into either eating or rumination activity.
From mean prediction error observed at the day scale, it can be extrapolated that at individual cow scale, at small herd size scale like in this study (4-7 cows), or at larger herd size scale (>20 cows), the accuracy of the FeedPhone seems sufficient to detect between-day variations of eating and rumination times of about 20, 10 and 5%, respectively. Additional validation studies are needed to determine accuracy of the FeedPhone on cows fed on pasture silage, hay and/or under grazing.
Conclusions
At the day level, the FeedPhone is able to record automatically the eating and rumination time of dairy cows fed on a maize silage-based diet with a precision of 89 to 90% for both activities. Data analysis at the hour scale shows that algorithms are specific enough for clearly identifying the eating and rumination activities. This overall accuracy is good enough to make the FeedPhone a valuable tool for studying relative variations of both eating and rumination times of dairy cows.
References
Baumont R., Chenost M. and Demarquilly C. (2004) Measurement of herbage intake and ingestive behaviour by housed animals.
In: Penning P.D. (ed) Herbage intake handbook. British Grassland Society, UK, pp. 121-150.
Bibby J. and Toutenburg H. (1977) Prediction and improved estimation in linear models. Wiley, London, UK.
Nielsen P.P. (2013) Automatic registration of grazing behaviour in dairy cows using 3D activity loggers. Applied Animal Behaviour Science 148, 179-184.
Umemura K. (2013) Monitoring grazing bites and walking activity with pedometers. Journal of Dairy Science 96, 1090-1093.
Table 1. Accuracy of the FeedPhone to record eating and rumination time in lactating dairy cows fed on a total mixed ration based on maize silage (89 d and 2,212 h of validation).
Scale Activity INRA
min
FeedPhone min
MPE
1MSPE
1min % mean % mean % slope % random
Day eating 364 346 41.8 11.5 18 27 55
rumination 531 523 58.8 11.1 2 0 98
Hour eating 14.6 13.9 5.7 39.2 2 1 97
rumination 21.4 21.1 7.5 35.2 0 3 97
1
